* feat: refactor ai to have single entry point and follow defined model structure
* fix: missing openrouter registration
* fix: dedupe models
* feat: provide usage in stream mode 🚀
We weren't able to see any custom spans, only the builtin
instrumentations for node.js. This commit downgrades opentelemetry so
that we can see custom spans in tools like jaeger as before.
Introduces Mistral OCR integration in backend and updates the `AIInterfaceService` and `MistralAIService` to support new OCR options and response normalization. Updates metering cost map for OCR and annotation usage. Refactors `DriverService` to support interface-specific service aliases. Expands the puter.js `AI.img2txt` API to support flexible options and provider selection, including Mistral OCR.
These two issues have affected us, but go away without rollup:
- https://github.com/npm/cli/issues/4828
- https://github.com/rollup/rollup/issues/6168
We can just use webpack instead; it has never caused build/npm issues.
This migration was done using AI, by pasting error messages over and
over again until everything worked. Since we're not actively working on
terminal and phoenix (dependents of rollup) currently it doesn't make
sense to focus on doing this migration manually, and any errors not
observed in basic use can be remedied later.
The styleguide for Puter's backend expects spaces after top-level not
operators inside conditions. This commit adds an AI-generated eslint
plugin and, since it conflicts with the existing space-unary-ops plugin
from stylistic, also adds a wrapped version of space-unary-ops that
ignores top-level not operators so that this plugin will work as
expected.
- A new stt driver in `AIInterfaceService` and `PuterAIModule`.
- added methods for audio transcription and translation in the speech-to-text interface.
- updated cost mapping for stt models in `openAiCostMap.ts`.
- Updated permissions and interfaces to support new speech-to-text features.
* Don't update URL when opening public files via direct URL
* Alert user when attempting to open a nonexistent file
* Fix toolbar autohide in public file urls
* Remove extraneous debug logs
* Properly define stat var in public folder URL handling
* Update window URL when focusing an app opened through file URL
* Localize error messages
* Update OpenAI package to version 6.7.0 and add video generation capabilities
- Updated `openai` dependency in `package-lock.json` and `package.json`.
- Introduced new permissions for video generation in `hardcoded-permissions.js`.
- Added video generation service in `AIInterfaceService.js` and registered it in `PuterAIModule.js`.
- Implemented coercion for video streams in `CoercionService.js`.
- Updated driver service to include video generation.
- Added `txt2video` method and options in `AI.js` and corresponding TypeScript definitions in `index.d.ts`.
* Update openaiVideoCostMap.ts
* Revert "Update openaiVideoCostMap.ts"
This reverts commit 9e66f93605.
* change `txt2video` to `txt2vid`
* Add 'Set as Desktop Background' context menu item
* Use the `mime` library to detect images in the `UIItem` context menu
---------
Co-authored-by: Nariman Jelveh <nj@puter.com>
Daniel Salazar